$492M
AI governance market 2026
Gartner, Feb 2026
Gartner, Feb 2026
$4.63M
Avg shadow AI breach cost
IBM 2025
IBM 2025
48%
CISOs: agentic AI is #1 attack vector
40:1
Non-human to human identity ratio in enterprise
The Insight
Every AI agent builds its own model of you. You own none of it.
Claude has its memory. ChatGPT has its memory. Copilot has its memory. You can't port it, can't see it, can't control it. In March 2026, Anthropic and Google launched memory import tools — racing to be the walled garden that owns your context. OpenAI won't even let you export.
~/.claude/CLAUDE.md — the artisanal prototype
# Patrick's Global Claude Preferences ## Who I am Founder and operator. Background in senior product roles. Comfortable with technical work but not a full-time engineer. ## How I work - I prefer to understand *why* before *what* - Don't over-engineer. Propose the simplest thing that works. - Ask 3-5 clarifying questions at a time ## Communication style - Be direct. Skip preamble and filler. - If I ask for a recommendation, give one — don't hedge.
Every AI power user maintains some version of this. It's artisanal, manual, and not portable. But it proves the need: people want to own their AI identity.
Product Architecture
Three Layers
🪪 Identity Vault (PID)
Portable, versioned profile: communication preferences, role context, workflow patterns, privacy rules. The artifact that travels between agents.
🔐 Authorization Engine
Per-agent scoping: what it can read, what it can do, what requires approval. Built on existing Totem delegation engine — verification, audit trails, revocation.
📊 Governance Surface
Cross-agent dashboard: see what every agent knows, what it's done, what's pending. Rule-based HIL management in the user's terms.
The HIL Innovation
Sharper than what agents offer natively
Native Agent HIL
Per-action: "approve this tool call"
Single agent, siloed
Agent's internal framing
Ephemeral yes/no clicks
Agent-defined options
Totem HIL
Rule-based: "can read docs, can't send emails"
Cross-agent, unified surface
User's terms, with context for why
Logged with full context
User-defined rules and policies
Competitive Landscape
The gap nobody fills
Developer Memory Infra
Mem0 ($24M) / Letta ($10M)
Developer-facing. User doesn't own or port memory.
Enterprise Agent Security
Okta / CyberArk / Aembit
Controls what agents reach, not what they know about you.
Portable Context Protocol
Plurality Network
Protocol-first, crypto-adjacent. Enterprise wants product.
Platform Memory
Claude / ChatGPT / Copilot Memory
Siloed. Each platform wants lock-in, not portability.
TOTEM
User-owned portable identity + enterprise governance door + product-first with protocol north star
Thesis Fit
"You Are an API" — AI agents are today's black boxes. They hold your context hostage, each builds its own model of you, and you have no portability or transparency. Totem extracts that identity, gives the user ownership, and makes the person the authoritative source agents call into.
Wherewithal Ventures — Business Model Inversion thesis
Business Model
Freemium wedge, enterprise revenue
Free
$0
Individual
- PID vault
- 2 agent connections
- Basic HIL rules
Pro
$12
per month
- Unlimited connections
- Advanced rules
- Full audit history
- Import tools
Enterprise
$8-15
per seat / month
- Policy templates
- Compliance dashboards
- Org-wide controls
- SSO + audit exports
Roadmap
Product → Protocol
Signal
Validate the insight
Document set, competitive scan, user conversations
Frame
Build v0 vault + auth
Claude via MCP, PAM as testbed
Build
Enterprise governance
Revenue, multi-agent, compliance
Prove
Protocol emergence
Open standard, ecosystem adoption
Pressure Test
Investor Review Board Assessment
Strengths
Insight is sharp and timely — March 2026 memory wars prove the pain
Competitive gap is real: between Mem0/Letta and Okta/CyberArk, nobody owns user identity
CLAUDE.md origin story is compelling — demand already proven by manual behavior
Existing delegation engine provides real technical head start
Regulatory tailwind: GDPR Article 20, EU AI Act transparency requirements
Risks to Watch
TAM is borrowed from adjacent markets — "agentic identity" category doesn't exist yet
Plurality Network is a real threat if they ship enterprise UX
Copy problem: access control ≠ information control (honest but limits privacy story)
Enterprise procurement may be slow — bottom-up adoption is the hedge
Solo founder without deep identity/security engineering background
The Ask
$500K
Seed via PSL
12 months runway
12 months runway
$2M
ARR target
Series A trigger
Series A trigger
18mo
To $2M ARR
3 enterprise customers + multi-agent
3 enterprise customers + multi-agent